
🤓 Can't Get Any Smarter? New AI Exam Ends in Embarrassment for Machines
Sounds like a plot from a sci-fi disaster movie. But it's reality: scientists have launched "Humanity's Last Exam" - a test designed to show just how close smart machines really are to us.
To understand the limits of AI, nearly a thousand researchers from around the world (historians, physicists, linguists, medical professionals) compiled 2,500 questions from real academic backwaters: from translating ancient Palmyrene inscriptions to distinguishing anatomical structures in birds.
The results are modest: the best models barely score 40-50% correct answers.
📌 Lifehack for checking competencies: If you suspect an AI "hallucination," try using the "Socratic dialogue" method.
"Explain step by step why you think so. What research are you relying on?" If the AI mumbles or references non-existent works, it's likely a typical "hallucination."
We humans can breathe a sigh of relief)
👉 Typical "AI hallucination" via link
Comments
0No comments yet.